From owner-sc22wg5@open-std.org Thu Apr 22 00:49:09 2010 Return-Path: X-Original-To: sc22wg5-dom8 Delivered-To: sc22wg5-dom8@www2.open-std.org Received: by www2.open-std.org (Postfix, from userid 521) id 4A8BFC4E9FA; Thu, 22 Apr 2010 00:49:09 +0200 (CET DST) X-Original-To: sc22wg5@open-std.org Delivered-To: sc22wg5@open-std.org Received: from ironport2.lbl.gov (ironport2.lbl.gov [128.3.41.14]) by www2.open-std.org (Postfix) with ESMTP id 1BC9AC3BA26 for ; Thu, 22 Apr 2010 00:49:06 +0200 (CET DST) X-Ironport-SBRS: None X-IronPort-Anti-Spam-Filtered: true X-IronPort-Anti-Spam-Result: ApsEAEIdz0uAAwU8/2dsb2JhbACdDL4cAoJyghoEgzQ X-IronPort-AV: E=Sophos;i="4.52,252,1270450800"; d="scan'208";a="108940432" Received: from angilas.lbl.gov (HELO angilas.localnet) ([128.3.5.60]) by ironport2.lbl.gov with ESMTP/TLS/DHE-RSA-AES256-SHA; 21 Apr 2010 15:49:04 -0700 From: Aleksandar Donev Organization: LBL To: WG5 Subject: Re: (j3.2006) (SC22WG5.4255) [ukfortran] Question on locks Date: Wed, 21 Apr 2010 15:49:04 -0700 User-Agent: KMail/1.11.2 (Linux/2.6.28-17-generic; KDE/4.2.2; x86_64; ; ) References: <20100421204947.D8CC7C178DA@www2.open-std.org> <20100421211846.2761BC4E9FA@www2.open-std.org> <20100421221958.90A94C4E9FA@www2.open-std.org> In-Reply-To: <20100421221958.90A94C4E9FA@www2.open-std.org> MIME-Version: 1.0 Content-Type: Text/Plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Content-Disposition: inline Message-Id: <201004211549.04174.adonev@lbl.gov> Sender: owner-sc22wg5@open-std.org Precedence: bulk On Wednesday 21 April 2010, N.M. Maclaren wrote: > What locks don't provide is > a barrier - i.e. a polint at which two images synchronise, and > everything before is before, and everything after is after. They > provide a one-way ordering. OK, Nick triggered my memory. He is right. The reason we took the "implied sync memory" is that locks provide one-way ordering, and allow for code-movement (i.e., reading register values or preloading etc.) in one direction, but not the other. This was specifically requested by some optimization people (from the C++ committee) and semantically much better anyway. So the answer to your question is "no, sync memory is not required in Note 8.44." As Nick mentioned, Note 8.40 uses atomics and it does require the sync memories. Aleks -- Aleksandar Donev, Ph.D. Luis W. Alvarez Postdoctoral Fellow Center for Computational Sciences and Engineering (https://ccse.lbl.gov) Lawrence Berkeley National Laboratory (http://www.lbl.gov) E-mail: adonev@lbl.gov Phone: (510) 486-5782 Fax: (510) 486-6900 Address: MS 50A-1148, LBL, 1 Cyclotron Rd., Berkeley, CA 94720 Web: http://cims.nyu.edu/~donev/